Neuron Explorer

Qwen3-0.6B

commas inside numbers and lists

A neuron in layer 12 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron reacts to commas that split up big numbers like 65,000 or place names like Newark, DE. When it is turned down, it is most active on commas that separate items in a list of actions. Turning it up did not produce a clear change in topic.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    . This one, called iBART, was a thing of beauty. Free, too. It was written by two former high-school buddies—Ian Leighton, a sophomore at UC Berkeley, and David Hodge, a sophomore at the University

  2. #2 peak

    pelier, Vermont, on December 26, 1837. A hero of the Civil War and Spanish-American War, he was the first, and to date only, person promoted to Admiral of the Navy.⏎Marihuana

  3. #3 peak

    five-a-day, it would now seem that even that is not enough.⏎Nutritional boffins at the University College London surveyed 65,000 people over 12 years and concluded that we should actually be

  4. #4 peak

    nervous system, causing many of the disease’s symptoms.⏎How common is Citrullinemia Type 1?⏎Scientists estimate that 1 in 57,000 births are affected by Citrullinemia type I.⏎How

  5. #5 peak

    really learn.⏎Beers, Kylene and Robert E. Probst. Notice & Note: Strategies for Close Reading. Portsmouth, NH: Henemann, 2013.⏎Using the work of Louise

  6. #6 peak

    three Stanford researchers that delves into those pitfalls appears in Proceedings of the National Academy of Sciences. For their work, the team analyzed the results of 50,000 studies in 22 fields of science. “I think that

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: I have been on the internet, it's been a lot of social media. I've been working in the dark hours of work, I have been on a lot of social media, I have been on a lot of work. I have

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    this product is for A SINGLE USER ONLY. That means that copying or sharing any part of this product with another teacher, grade level, school, department, district,

  2. #2 peak

    , engines, or generators to control and adjust production of biofuels or biomass-fueled power.⏎- Operate equipment to start, stop, or regulate biomass-fueled generators, generator units, boilers, engines,

  3. #3 peak

    ||fortran|⏎This entry (in part or in whole) was last reviewed on 4 February 2012.⏎Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version

  4. #4 peak

    ctman & Horowitz, 2003). Studies find that people do not treat computers as people outside the mediation context (Goldstein, Alsio, & Werdenhoff, 2002) – just as people do

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L10:667 suppresses commas inside lists of names
  2. L10:74 suppresses
  3. L6:620 suppresses "and" in lists of three or more
  4. L8:1605 excites
  5. L9:77 excites commas in citations and name lists
  6. L9:738 suppresses
  7. L7:866 suppresses
  8. L7:288 excites
  9. L11:154 suppresses commas after citation numbers
  10. L8:2014 excites
  11. L8:1046 weakly excites
  12. L8:1652 weakly suppresses
  13. L10:1953 weakly suppresses
  14. L11:807 weakly suppresses
  15. L3:661 weakly excites
  16. L10:2676 weakly suppresses
  17. L4:3 weakly excites commas after citation fields
  18. L9:633 weakly excites
  19. L5:16 weakly excites
  20. L11:876 weakly excites 'other' after 'among' in lists
  21. L0:2986 weakly suppresses
  22. L0:2436 weakly suppresses
  23. L9:509 weakly suppresses
  24. L10:1643 weakly excites

Feeds later layers

  1. L13:977 strongly suppresses
  2. L13:758 suppresses 'and' joining a pair of options
  3. L16:1388 suppresses
  4. L14:1054 suppresses
  5. L15:677 suppresses
  6. L13:517 suppresses comma closing a list item
  7. L14:874 suppresses
  8. L15:611 suppresses
  9. L13:125 suppresses
  10. L14:1804 suppresses "and" in subject-area lists
  11. L15:545 suppresses
  12. L21:735 suppresses 'or' in ranges and alternatives
  13. L17:572 suppresses
  14. L16:1294 suppresses
  15. L14:2054 suppresses
  16. L14:1215 suppresses
  17. L16:2733 suppresses
  18. L16:1234 suppresses
  19. L14:2961 suppresses
  20. L13:900 weakly suppresses
  21. L19:1369 weakly suppresses
  22. L16:774 weakly suppresses
  23. L14:1746 weakly suppresses
  24. L15:631 weakly suppresses

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

1 values between -7.05 and -6.851 values between -6.64 and -6.441 values between -5.02 and -4.821 values between -4.00 and -3.801 values between -3.80 and -3.601 values between -3.39 and -3.192 values between -3.19 and -2.992 values between -2.99 and -2.782 values between -2.78 and -2.583 values between -2.58 and -2.384 values between -2.38 and -2.181 values between -2.18 and -1.971 values between -1.97 and -1.776 values between -1.77 and -1.571 values between -1.57 and -1.367 values between -1.36 and -1.164 values between -1.16 and -0.967 values between -0.96 and -0.7517 values between -0.75 and -0.5528 values between -0.55 and -0.35281 values between -0.35 and -0.142192 values between -0.14 and 0.061195 values between 0.06 and 0.26128 values between 0.26 and 0.4714 values between 0.47 and 0.672 values between 0.67 and 0.875 values between 0.87 and 1.08offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. inction
  2. .Dict
  3. owing
  4. either
  5. แดน
  6. 既
  7. 大片
  8. .ByteString
  9. umbing
  10. 初

Suppresses

  1. reas
  2. uku
  3. rine
  4. quito
  5. 和他的
  6. aches
  7. ことです
  8. chas
  9. aru
  10. lement